Can I use just one model to solve multiple multiclass problem?


I am currently working with a problem recognizing multiple color labels of a urinalysis strip.

There are 10 different chemical indicators (The first one on the left is just for color correction) on the strip to test different aspects of urine. Each indicator has 3 to 7 colors which represent the reading of it.

So it’s a multilabels-multiclasses problem.

Instead of training 10 models to recognize the color of each of the indicators, is there any better solution?

I would be grateful for any suggestion! Thank you!

You could train a multiclass model, but instead of softmax use sigmoid as final activation function (or no final activation at all). CE Loss should work nonetheless, but you need different metrics for multi label classification.

1 Like